Start ecommerce CRO with a question a buyer cannot answer or an action they cannot complete. Show the right size, explain the total, fix the variant selector or remove an overlay that blocks the cart. Then test a specific improvement against an outcome you defined in advance. A new button color is useful only when there is a reason to believe it addresses the problem.
A fictional shoe page with a fit problem
Imagine a shoe store receiving repeated questions about whether its sizes run small. The product page has attractive photographs but no verified fit guidance. The team proposes a bigger discount. A more direct first project is to gather dependable sizing information and make it easy to use.
The hypothesis is specific: buyers hesitate because they cannot confidently choose a size. The proposed change is a clear size guide with verified measurements and an explanation of how to measure. That is a better brief than “improve the page.”
This is a fictional planning example. It does not establish that sizing is the cause at your store or that adding a guide will raise sales. Support questions, return reasons and a real buying walkthrough should guide your choice.
Separate a repair from an experiment
If selecting a size adds the wrong variant to the cart, repair it and verify the corrected behavior. You do not need to preserve a broken experience for half the visitors to prove it is a problem. Check mobile and desktop, unavailable variants and returning to the cart.
If two clear, functional explanations might work differently, an experiment can help. Name the intended audience, primary outcome, observation window and guardrails before starting. Keep refund or return effects visible when a change could persuade people to buy the wrong product.
Use the product-page audit to find purchase blockers before planning a series of cosmetic tests.
Know what the arithmetic does and does not say
Fictional illustration: 10,000 eligible sessions produce 200 orders, an order-per-session rate of 2%. Another 10,000 produce 230 orders, or 2.3%. That is 30 more orders and a 0.3 percentage-point difference.
The arithmetic alone does not prove the page change caused the difference. Allocation, sample size, repeat visits, traffic mix and other changes affect the interpretation. A before-and-after comparison is especially vulnerable to promotions or stock changes. Do not announce a winner because the first afternoon looks encouraging.
Decide whether the available traffic can answer the question with useful confidence. When it cannot, prioritize reproducible usability defects and clearer information, and label the commercial effect unproven.
Finish with a usable experiment note
Record the observed problem, evidence, exact change, eligible audience, metric definition, dates and result. Add what else changed during the test. Save screenshots and the rollback version.
Read results alongside contribution and returns. More orders with a larger discount can still leave less contribution. Your next step is to watch one real buying path on a phone and write down the first unresolved purchase question. Make that the next brief.
Editorial revision, October 20, 2026: unsupported conversion lifts, tool pricing and universal performance benchmarks have been removed. Examples are fictional and no improvement is guaranteed.